laion/majestrino-1.00-16xk5-sae-features

---
license: apache-2.0
task_categories:
- audio-classification
language:
- multilingual
tags:
- sparse-autoencoder
- voice
- audio
- interpretability
- majestrino
pretty_name: "Majestrino 1.00 SAE Feature Audio Samples"
size_categories:
- 1M<n<10M
---

# Majestrino 1.00 SAE — Feature Audio Samples (16x, k=5)

Top-2000 activating audio samples for each feature in the
[Majestrino 1.00 SAE](https://huggingface.co/laion/majestrino-1.00-16xk5-sae).

## Overview

| Metric | Value |
|--------|-------|
| SAE Architecture | 16x expansion, k=5, d_model=768 |
| Total Features | 12,288 |
| Alive Features | 10,684 |
| Audio per Feature | Up to 2,000 highest-activating |
| Audio Format | Opus (24 kbps OGG container) |
| Total TAR Files | 1069 |
| Source Dataset | [laion/majestrino-data](https://huggingface.co/datasets/laion/majestrino-data) |

## File Structure

Each TAR file contains 10 features, named `features_XXXXX_YYYYY.tar`.

Inside each TAR:
```
feature_00042/
metadata.json # Feature info, annotation, activation scores
00001991.opus # Audio file (highest activation)
00002299.opus # Audio file (2nd highest)
...
feature_00043/
metadata.json
...
```

### metadata.json

```json
{
"feature_id": 42,
"title": "British Male Narrator",
"description": "This feature activates on...",
"bin": 12,
"bin_name": "Broadcast & Formal Style",
"consistency": 3,
"activation_count": 15234,
"n_audio_files": 2000,
"activations": [
{"file": "00001991.opus", "activation": 0.8234, "original_path": "...", "tar_source": "00042"},
...
]
}
```

## Usage

```python
import tarfile, json

# Extract a feature TAR
with tarfile.open("features_00000_00009.tar") as tf:
tf.extractall("./extracted")

# Read metadata
with open("./extracted/feature_00042/metadata.json") as f:
meta = json.load(f)

print(f"Feature {meta['feature_id']}: {meta['title']}")
print(f"Top activation: {meta['activations'][0]['activation']:.4f}")
```

## Related

- **SAE Model**: [laion/majestrino-1.00-16xk5-sae](https://huggingface.co/laion/majestrino-1.00-16xk5-sae)
- **Source Data**: [laion/majestrino-data](https://huggingface.co/datasets/laion/majestrino-data)